Distributed photovoltaics (DPV) are a growing source of electricity generation in the United States, and with adoption driven by customer behavior and localized economics, projecting the deployment of this technology is a challenging analytical problem. Moreover, understanding the sources of uncertainty in customer adoption models and how they can be reduced is important to a range of stakeholders that use their outputs, including grid planners, regulators, and industry. Most prior studies have used top-down methods, such as the use of population central tendencies to project aggregate adoption. In contrast, a growing field of work seeks to use bottom-up methods (i.e., individual-level decision-making).We explore trade-offs of top-down and bottom-up methods in their precision and computational burden using the National Renewable Energy Laboratory's (NREL's) Distributed Generation Market Demand (dGen) model, an agent-based model of residential and nonresidential distributed PV adoption. In particular, we assess the role of agent resolution in instantiating statistically-representative populations in the model-and the resulting variance of model projections at the state, sector, and county levels. At low sampling rates, the model resembles a top-down model, whereas as sampling rates increase dGen converges to a bottom-up structure by simulating more unique customer types. Though sampling-based models such as dGen can be operated with many agents to ensure accuracy, doing so greatly increases the computational burden of the simulation. This report lends insight into whether high-resolution results can be approximated sufficiently well using fewer computational resources.vii